Search

BackgroundFederated learning is a decentralized approach to machine learning; it is a training strategy that overcomes medical data privacy regulations and generalizes deep learning algorithms. Federated learning mitigates many systemic privacy risks by sharing only the model and parameters for training, without the need to export existing medical data sets. In this study, we performed ultrasound image analysis using federated learning to predict whether thyroid nodules were benign or malignant. ObjectiveThe goal of this study was to evaluate whether the performance of federated learning was comparable with that of conventional deep learning. MethodsA total of 8457 (5375 malignant, 3082 benign) ultrasound images were collected from 6 institutions and used for federated learning and conventional deep learning. Five deep learning networks (VGG19, ResNet50, ResNext50, SE-ResNet50, and SE-ResNext50) were used. Using stratified random sampling, we selected 20% (1075 malignant, 616 benign) of the total images for internal validation. For external validation, we used 100 ultrasound images (50 malignant, 50 benign) from another institution. ResultsFor internal validation, the area under the receiver operating characteristic (AUROC) curve for federated learning was between 78.88% and 87.56%, and the AUROC for conventional deep learning was between 82.61% and 91.57%. For external validation, the AUROC for federated learning was between 75.20% and 86.72%, and the AUROC curve for conventional deep learning was between 73.04% and 91.04%. ConclusionsWe demonstrated that the performance of federated learning using decentralized data was comparable to that of conventional deep learning using pooled data. Federated learning might be potentially useful for analyzing medical images while protecting patients’ personal information.

Background Methods Results Conclusions Papillary thyroid carcinoma (PTC) occasionally invades the trachea and requires airway resection. Tracheal excision site recurrence (ESR) is a serious problem. We investigated predictors of ESR in patients with PTC who underwent airway resection for locally curative surgery.We enrolled 149 patients with PTC who underwent airway resection (median age at the initial surgery: 67 years), including partial‐thickness resection (n = 73) or full‐thickness resection (n = 76), for grossly curative surgery. The median postoperative follow‐up period was 93 months.To date, 11 patients (6.7%) underwent ESR: 6 underwent full‐thickness resection and 5 underwent partial‐thickness resection. The time to ESR ranged from 14 to 113 months (median: 57 months) after the initial surgery. None of the 11 ESR patients underwent adjuvant external beam radiotherapy (EBRT) and none of the 4 airway resection patients who underwent EBRT developed ESR. The 5‐ and 10‐year ESR rates were 4.3% and 11.3%, respectively. In the multivariate analysis (forward–backward stepwise selection method), a Ki‐67 labeling index (LI) ≥5% (p = 0.048) and the thyroglobulin doubling rate (Tg‐DR) >0.33/year (p = 0.009) (for Tg‐antibody negative cases) were independent predictors of ESR. Nine of the 11 patients underwent ESR resection and only one developed a second recurrence.A high Ki‐67 LI was a static predictor, and high Tg‐DR was a dynamic predictor, of ESR in patients with PTC following airway resection. In such patients, careful postoperative monitoring for ESR is necessary and adjuvant therapies, such as EBRT, may be considered. [ABSTRACT FROM AUTHOR]

Objectives: Injury to the external branch of the superior laryngeal nerve (EBSLN) causes low‐pitch voice and voice fatigue, particularly in female subjects, and available treatments are limited. Here, we assess a novel surgical procedure to restore a high‐tone voice: ansa cervicalis to EBSLN anastomosis (A‐E anastomosis). Methods: Between November 2012 and April 2022, 13 patients (12 female) underwent unilateral EBSLN resection and A‐E anastomosis, while 20 (16 female) underwent EBSLN resection during thyroid surgery. Patients (4494 women and 1025 men) with normal laryngoscopy scheduled for thyroid surgery served as normal controls. Phonatory function was examined using a Phonation Analyzer PA‐1000 preoperatively and intermittently postoperatively. Results: In patients who underwent A‐E anastomosis, high‐tone voice pitch decreased significantly postoperatively (673.9–471.5 Hz, p = 0.047), with restoration achieved within 5 months. The mean voice pitch in female patients who underwent A‐E anastomosis, EBSLN resection, and controls were 580.4, 522.8, and 682.0 Hz, respectively, indicating a significant decrease in EBSLN resection patients than controls (p = 0.002). The (mean – 1SD) of high‐tone voice pitch in female controls was 497 Hz; exceeding this may indicate recovery to a high‐tone voice. Overall, 73% (8/11) of A‐E anastomosis patients exceeded this value, which was marginally larger than the 43% (6/14) who underwent EBSLN resection. Data on male subjects are limited. There were no cases of adverse functional or cosmetic events. Conclusions: A‐E anastomosis, a novel simple procedure, restored high‐tone voice to some extent without any adverse events and thus warrants further investigation. Level of Evidence: 3 Laryngoscope, 134:3868–3873, 2024 [ABSTRACT FROM AUTHOR]

Background Methods Results Conclusions Papillary thyroid carcinoma (PTC) often extends to adjacent organs. According to the 8th Tumor‐Node‐Metastasis Classification, extension to the strap muscles was graded as T3b. We investigated the prognostic impact of T3b and the appropriateness of T3b in patients aged ≥55 years who were classified as stage II.We enrolled 7811 patients with M0 PTC who underwent initial surgery at the Kuma Hospital (Kobe, Japan) between January 2007 and December 2016. Tumor extension was divided into T3b, T4a1 (extension to the tracheal adventitia, tracheal cartilage, esophageal muscle layer, recurrent laryngeal nerve, cricothyroid, and inferior constrictor muscles), and sT4a2 (extension to the subcutaneous soft tissues, tracheal mucosa, esophageal mucosa, internal jugular vein, brachiocephalic vein, larynx, pharynx, and sternocleidomastoid muscle).In patients ≥55 years, the local recurrence‐free survival (LR‐FS), distant recurrence‐free survival (DR‐FS), and cause‐specific survival (CSS) rates of T3bN0M0 were significantly poorer than those of T1/T2N0M0 but did not significantly differ from those of T3aN0M0. The LR‐FS, DR‐FS, and CSS rates of T3b stage II patients did not differ from those of T4a1 stage III patients but were significantly better than those of T4a2 stage III patients. T3b was an independent predictor of local recurrence and distant recurrence but not of death due to carcinoma in the multivariate analysis. In patients aged <55 years with M0 PTC, T3b had no prognostic value in both analyses.T3bM0 patients are appropriate to be classified as stage II in patients ≥55 years but be kept in stage I in patients <55 years. [ABSTRACT FROM AUTHOR]

Objectives: The purposes of this study were to validate the Acoustic Breathiness Index (ABI) for the Japanese-speaking population and to determine whether it is independent of factors such as sex, age, and perceptual ratings of roughness. Method: First, the concurrent validity of the ABI for perceptual breathiness was evaluated on the concatenations of continuous speech and sustained vowels from 288 patients with varying degrees of dysphonia. The diagnostic accuracy was examined on 343 samples with 55 additional normophonic speakers. Second, the validity related to responsiveness-to-change was estimated on 222 samples obtained before and after interventions for 111 voice-disordered patients. Third, the relationships between the ABI and other variables (i.e., perceptual hoarseness/breathiness/roughness, sex, and age) were explored using bivariate and multivariate analyses for the 288 patients. Results: First, the concurrent validity and the responsiveness-to-change validity were confirmed by strong correlation coefficients of 0.890 and 0.878, respectively. Second, the receiver operating characteristic analysis showed the area under the curve to be 0.939, indicating excellent accuracy. The ABI of 3.44 exhibited a sensitivity of 76.3% and a specificity of 94.1%. Third, although bivariate analyses revealed a weak relationship between ABI and roughness and an ABI difference by age, multiple regression analyses showed a strong relation between only ABI and breathiness, without a meaningful contribution from roughness, sex, and age factors. Conclusion: The study confirmed that the ABI is an accurate and specific tool to estimate breathiness levels in the Japanese-speaking population and neither roughness, sex, nor age significantly affects the ABI.

Objectives: The acoustic voice quality index (AVQI) is a reliable tool that objectively assesses dysphonia levels using six acoustic parameters. Despite its high criterion‐related concurrent validity, diagnostic accuracy, and minimal detectable change derived from test–retest reliability, the minimal important difference (MID) of the AVQI has not been tested before. This study aimed to estimate the MIDs of AVQI for improvement audibly perceived by clinicians and self‐reported improvement by patients. Methods: A retrospective study was conducted on 110 patients who received treatment for voice disorders. Patients completed AVQI and Voice Handicap Index‐10 (VHI‐10) questionnaires before and after the therapy. The MIDs of the AVQI were estimated using the anchor of either auditory–perceptual judgment of total dysphonia levels by clinicians or the VHI‐10 questionnaire by patients. A distribution‐based approach was also used to complement the results. Results: First, using the auditory–perceptual anchor, a decrease of 0.95 in the AVQI was estimated as the MID for clinicians' perception, as a result of the receiver operating curve. Then, using the patient‐reported anchor, an improvement of 1.36 in the AVQI was estimated as the MID for patients' voice‐related disability. The distribution‐based approach also ensured the anchor‐based results of both the MIDs. Conclusions: The AVQI is a reliable and valid tool for evaluating voice quality, and a 0.95 decrease in the AVQI represents a meaningful improvement for clinicians' perception, whereas a 1.36 decrease in the AVQI influences patients' self‐reported disability. This study contributes to understanding the minimal change necessary for clinicians to make informed decisions and ensure patient satisfaction. Level of Evidence: 3 Laryngoscope, 134:2805–2811, 2024 [ABSTRACT FROM AUTHOR]